Electroencephalography-Based Motor Imagery Classification Using Multi-Scale Feature Fusion and Adaptive Lasso
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe topic on “EEG-based motor imagery classification using multi-scale feature
fusion and adaptive lasso” is gaining interest both in the medical and the engineering filed.
While the manuscript is interesting, there are some aspects that should further be addressed:
1. How do you plan to adapt the usage of your method for the hemiparetic patients for example? It would have been interesting to also have a group of stroke survivors or patients with any CNS injury to the impact of adapting the method.
2. The classical motor imagery could be also applied for another group, while the other is using the EEG-based motor imagery. Did you gather these data as well? If so, it would be interesting to add it.
3. It is not entirely clear to me how easily can this work be applied to the classical EEG machines and adapting to the sometimes lacking complementary information on some of the features, notably the spatial ones.
4. Given that you stated that the proposed MIC-Lasso helps to identify and adapt the most discriminative EEG features, hence enhancing MI-EEG decoding accuracy, for future research I recommend a multidisciplinary team where patients and several neurologic pathologies should be included.
Good luck!
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe proposed machine learning framework for motor imagery EEG classification presents a significant advancement in the field. However, a more comprehensive examination of its practical applicability, such as its performance in real-time or online brain-computer interfaces, could enhance its relevance. Moreover, the dependency on benchmark datasets may restrict its generalizability to more diverse or noise-prone real-world EEG data. The computational complexity of the proposed framework, which is not thoroughly addressed, could also pose a challenge for practical implementation. Further discussions and a concluding remark are necessary to complete this article.
Abstract:
The abstract should clearly state the outline the study's purpose and motivation by highlight the significance of classifying motor imagery EEG signals, particularly for real-world uses like brain-computer interfaces (BCIs).
Also, the technical jargons like "multi-scale feature fusion" or "adaptive Lasso" should be avoided or briefly explained. Avoid too much technical details by summarizing key components of proposed methodology (Line 10-13).
Specify the quantitative results in terms of improvements in accuracy or other performance metrics (Line 13-16). Focus on how the proposed ML framework outperforms existing solutions or solves specific challenges in EEG classification.
At last end with a concluding remark that wraps up the contribution and importance of the work.
Introduction:
While reading the first paragraph (Line 20-39), a broader context to the significance of brain-computer interfaces (BCIs) and the role of electroencephalography (EEG) is missing from reader’s point of view.
Clearly state the motivations and challenges in motor imagery EEG classification. Explain the core issues in details such as inter-subject variability, low signal-to-noise ratio (SNR), and non-linear characteristics of EEG signals. This will justify why this study is necessary. The ML framework will work in back-end. It is the impact of this work that will support the importance of this work.
Line 40-107: Condense these paragraphs by a brief review of the current state of the art in EEG classification, focusing on existing methods such as common spatial pattern (CSP), deep learning models, and feature selection techniques. Emphasize where they fall short, especially in terms of handling variability and noise, and their limitations in subject-specific classification.
Line 108-136: Mention the key innovations of your approach (multi-scale feature fusion and adaptive Lasso) and briefly describe why it is effective.
adaptive Lasso) and briefly describe why it is effective.
Ensure that the introduction remains focused. Avoid too much technical details in this section, as these can be discussed in the later sections. The introduction should provide enough information to help readers understand the problem, existing solutions, and the contribution of your work without going into unnecessary depth.
In the subsequent section, “Related work”, they are discussed in detail and sounds like they are redundant.
Related Work:
The prior existing works are explained and summarized well; however, there is a lack of explicit comparison on how the proposed method improves upon or differs from existing methods. A few sentences explicitly stating a strong argument for why the existing methods fall short and why the proposed method is novel should be included. Additionally, incorporate content from the end-user perspective, emphasizing real-world applications.
Data Description:
Provide a more comprehensive description of the dataset characteristics, including participant demographics and any preprocessing steps applied to the datasets used (BCIC-IV-2a, SMR-BCI, OpenBMI).
Results and Discussion:
The sub-section 4.1 should be moved to Materials and method section as this is not a result.
While various classifiers such as KNN and SVM are discussed, providing a rationale for the selection of these specific classifiers that would enhance the clarity of the study. For example, elucidating the conditions under which KNN demonstrates superior performance in certain datasets. A more detailed comparison of their performance in relation to dataset characteristics would further strengthen the findings.
The ablation studies provide valuable insights but could be further enhanced by a more detailed analysis of how each feature fusion and selection mechanism affects performance. A clearer delineation of the impact on specific metrics, supported by additional tables or graphs, would benefit the comprehensiveness of the study.
Incorporating discussions on the errors or challenges encountered during classification and proposing potential solutions can provide valuable insights into areas for future enhancement. For instance, addressing limitations such as handling noise, overfitting in smaller datasets, or constraints related to computational resources could offer readers a deeper understanding of possible improvements.
Conclusions and Future Scope
Add discussion on potential real-world applications of the proposed methodology, such as BCI devices or medical diagnostics by emphasizing the practical impact of improved motor imagery classification.
For the future scope, elaborate more on the next steps, such as exploring more datasets, addressing challenges with individual variability in EEG, or improving model generalization for clinical use.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf